Inspiration
Traditional commerce platforms treat payments as passive endpoints: static links, manual invoice generators, or slow email reminders. Freelancers, agencies, and small businesses waste dozens of hours every month manually tracking deliverables, reconciling ledger accounts, and chasing overdue invoices.
Most recent "AI in fintech" implementations are merely conversational chatbots—they answer FAQs or tell you to visit a checkout link. We built PayPilot AI as a true Autonomous Action Agent: an intelligent financial operator that parses conversational goals in plain English, plans payment structures, enforces 5-point transaction safety checks, and manages the entire transaction lifecycle on the PayPal Developer Platform.
What it does
- Conversational Goal to Execution: Users enter natural language intents like "Collect $1,200 from Sarah for the website project by Friday". PayPilot extracts structured entities, verifies past customer ledger history, and provisions a structured payment plan.
- PayPal Developer Platform Integration: Directly integrates with PayPal Orders v2 REST API (
/v1/oauth2/token,/v2/checkout/orders,/v2/checkout/orders/{id}/capture), supporting live Sandbox execution and a transparent, deterministic in-app simulation lifecycle. - Payment Safety Engine: Protects against blind financial execution with 5 automated checks:
- $2,000 dual-authorization human review boundary.
- 24-hour duplicate payment velocity warnings.
- Strictly gated vendor disbursements (approvals record internal sign-off without auto-dispatching funds).
- Proactive Financial Intelligence: Recommends next actions (e.g., following up on overdue customer balances, highlighting pending review approvals).
- Durable Local Ledger: Maintains an immutable audit trail and reconciles customer account balances in real-time upon confirmed capture.
How we built it
- Frontend & UX: Built with Next.js 14 App Router, TypeScript, and Tailwind CSS, engineered with responsive 2-column cockpits and native 100% zoom layout.
- AI Planning Engine: Multi-agent planning powered by Google Gemini 2.5 Flash via official
@google/genaiSDK, backed by a deterministic NLP fallback for guaranteed offline reliability. - Payment Infrastructure: PayPal Orders v2 REST APIs with client credentials OAuth 2.0 and truth-in-advertising simulation fallback.
- Test Integrity: 46 automated unit and integration tests built with Vitest and Chai, enforcing rigorous regression protection and security boundaries.
Challenges we ran into
- Deterministic Action Gating: Ensuring an AI planner never accidentally triggers unvetted payouts or double-processes transactions. We solved this by decoupling AI planning from fund execution through an authoritative dual-authorization safety engine.
- Truthful Submission Disclosures: Designing an in-app simulation workflow that faithfully reflects PayPal's Orders v2 schema while being completely transparent with evaluators regarding buyer approval states.
Accomplishments that we're proud of
- 46/46 Vitest integration and security tests passing with zero failures.
- Zero-drift dual-engine architecture: verified live Google Gemini AI with resilient deterministic offline fallbacks.
- Fully accessible, interactive 1080p Full HD demo video under 2.5 minutes strictly conforming to hackathon rules.
What we learned
How to architect agentic workflows where LLM reasoning is safely constrained by deterministic financial rules, and how to build production-grade, state-persisted payment tools on top of PayPal's Orders v2 developer APIs.
What's next for PayPilot AI
- Real-time PayPal Webhook listeners (
CHECKOUT.ORDER.APPROVED,PAYMENT.CAPTURE.COMPLETED). - Native PayPal Invoicing API and Payouts v1 automated batch disbursements.
- Multi-currency intelligent hedging and scheduled milestone escrow releases.
Built With
- gemini-api
- google-genai
- next.js
- node.js
- paypal
- paypal-orders-v2
- react
- tailwind-css
- typescript
- vitest
Log in or sign up for Devpost to join the conversation.